Stipe, et. al,
I'm running this patch now (thanks for committing it). It looks like there is a slow leak of file handles in CLOSE_WAIT state. Is there any sort of connection reaping going on to find and remove these? (It doesn't appear so.) I believe the keep-alive patch helped a lot, but this leak is going to still run me out of file handles eventually. The file handles in question appear to belong to wapbox. I am running the latest CVS (obviously) of Kannel, with keep-alive disabled. It is doing http-proxy to Apache 2.0.52 on the same machine. Apache has keep-alive disabled as well as I'd mentioned before. I'm losing around 10 file handles (permanently) per minute I'd say, although that may be a bit inaccurate.
# netstat -np | grep -F CLOSE_WAIT | grep -F wapbox | wc -l 115
If you want me to check anything, I'll be more than happy to. Just let me know. I think I've got some time before I run out of file handles here and have to restart. Thanks. :)
Update: I've been watching both the wapbox log and the CLOSE_WAIT file handles, and it looks like it may happen on error conditions. Things like...
2005-03-01 09:59:34 [28652] [6] INFO: WTP_RESP: resp_machine_find_or_create: abort received, yet having no machine
2005-03-01 09:59:36 [32614] [0] WARNING: WTP PDU with unknown type -1
2005-03-01 09:59:36 [32614] [0] ERROR: WTP: cannot unpack pdu, creating an error pdu
...seem to add a permanent file handle to my list. I get a fair number of these for various reasons, and when I'm not getting them I get periods where my file handles stays static (which also makes sense). So it essentially looks like the error cleanup doesn't free the file handles in wapbox. HTH. :P
Jon
